1. STAR JPSS 2015 Annual Science Team Meeting
Enterprise Cloud Base
VIIRS Cloud Base Height Algorithm Improvement and Evaluation Using CloudSat
Yoo-Jeong Noh, John Forsythe, Curtis Seaman, Steve Miller, Matt Rogers
Colorado State University/CIRA
Dan Lindsey, Andy Heidinger
NOAA/NESDIS/Satellite Applications and Research

2. Introduction CBH (Cloud Base Height) is important for aviation.
Cloud ceiling and visibility - critical to the general aviation community
CBH is also important for closure of the Earth’s radiation budget in climate modeling.
CBH helps improve Cloud Cover Layer products.
A few attempts:
Hutchison (2002) developed algorithm to determine cloud base height (CBH) from VIS/IR observations from MODIS.
Chakrapani et al. (2002) and Minnis et al. (2005) developed CBH empirical parameterizations from GOES and ARM data.
We have been working on VIIRS CBH CAL/VAL and improvement using CloudSat data.
Herzegh et al. (BAMS 2015): DATA FUSION ENABLES BETTER RECOGNITION OF CEILING AND VISIBILITY HAZARDS IN AVIATION

3. Matching VIIRS with CloudSat
1353 UTC on 26 Sept 2013
S-NPP VIIRS True Color image
CloudSat CPR reflectivity
S-NPP
VIIRS CBH [km] with CloudSat overpass track (red)
from UTC on 26 Sept 2013
Suomi-NPP and CloudSat are in the same orbital plane, but at different altitudes
CloudSat and VIIRS overlap for ~4.5 hours every 2-3 days (8-9 matchups per month)
Due to battery issues, CloudSat only operates on the daytime side of the Earth
Use only the closest VIIRS pixels that overlap CloudSat and have CBH above 1 km
Parallax-corrected

4. VIIRS IDPS CBH Algorithm
VIIRS IDPS CBH algorithm for liquid clouds:
Red variables come from upstream retrievals.
LWC is pre-defined average value based on the upstream cloud type retrieval.
CBH for ice clouds is similar (T-dependent IWC). ,
VIIRS IDPS (colored) vs. CloudSat (gray)
CBH requires upstream retrievals of cloud properties which issues directly impact CBH retrieval.
As part of the JPSS Cloud Cal/Val efforts, our evaluation showed the IDPS CBH algorithm provided only marginal skill.
The cloud base height for liquid clouds is defined at right. Cloud base height definition for ice clouds is similar, except the average ice water content is temperature dependent. It was found that even a simple benchmark of 2 km geometric thickness would out-perform the EDR in a statistical sense.
Sample comparison results are shown at right. CloudSat Cloud Mask (gray, from 2B-GEOPROF) with VIIRS overlaid (IDPS CTH/CBH colored by cloud type)

5. Enterprise CBH Algorithm
CIRA’s new statistical Cloud Base Height Algorithm using A-Train satellite data
0-2 km top
MODIS Cloud Water Path (kg m-2)
Upper Layer Thickness (km)
12-14 km top
MODIS Cloud Water Path (kg m-2)
Upper Layer Thickness (km)
8–10 km top
Upper Layer Thickness (km)
MODIS Cloud Water Path (kg m-2)
Linear regression fits were performed between water path and geometric thickness for cloud top heights residing in 2 km vertical bins up to 20 km.
The median CWP value in each 2 km CTH bin was determined, and a linear regression above and below this value was performed.
An initial two-piece linear regression was performed for July daytime data from (1743 CloudSat/CALIPSO granules).
Cloud geometric thickness of the uppermost layer from the combined CloudSat/CALIPSO cloud profile product (2B-Geoprof-Lidar) and MODIS Cloud Water Path (MOD06)

6. Appropriate CIRA cloud geometric thickness (ΔZ) regression
Enterprise (Uppermost Layer) Cloud Base Data Flow
VIIRS CLAVR-x Cloud Type to modify cirrus and overshooting top bases
Extinction method for high thin cirrus
CBH=CCL_NWP for deep ‘overshooting’ type cloud
VIIRS CLAVR-x Cloud Water Path
CWP from COT and EPS using DCOMP and NLCOMP
CWP from NWP as supplementary data
VIIRS CLAVR-x Cloud Top Height
Appropriate CIRA cloud geometric thickness (ΔZ) regression
CBH retrieval
CBH = CTH - ΔZ

7. Investigating a Switch of Algorithms IDPS vs
Investigating a Switch of Algorithms IDPS vs. Enterprise CBH: “All Clouds”
The original IDPS with CLAVR-x input CIRA Statistical Regressions
show the preliminary comparison results between the original IDPS and the new statistical regression CBH algorithms when compared against CloudSat ‘truth’ observations for September – October 2013 matchups. The new CBH algorithm (beta version) has been derived based on a statistical linear regression between cloud top height and liquid/ice water path using CloudSat/CALIPSO and MODIS observations as described above. Initial results show the current statistical regression method outperforms the original IDPS CBH algorithm with CLAVR-x upstream input, with a tighter clustering of points along the 1:1 agreement line.
“All Clouds” evaluation: all clouds observed by CloudSat and VIIRS for the general performance
1540 VIIRS granules and matchup points for Sept-Oct 2013 cases
CBH [km]
Avg error (bias)
RMSE
Std of error r2
IDPS
1.0
3.3
3.1
0.286
Enterprise
3.0
2.8
0.427
Better

8. The enterprise CBH performs better. IDPS vs
The enterprise CBH performs better! IDPS vs. Enterprise CBH: “Within Spec”
The original IDPS with CLAVR-x input CIRA Statistical Regressions
show the preliminary comparison results between the original IDPS and the new statistical regression CBH algorithms when compared against CloudSat ‘truth’ observations for September – October 2013 matchups. The new CBH algorithm (beta version) has been derived based on a statistical linear regression between cloud top height and liquid/ice water path using CloudSat/CALIPSO and MODIS observations as described above. Initial results show the current statistical regression method outperforms the original IDPS CBH algorithm with CLAVR-x upstream input, with a tighter clustering of points along the 1:1 agreement line.
“Within Spec” evaluation for only clouds where the VIIRS CTH retrieval is within the error specifications: CTH within 1 km of CloudSat CTH if COT > 1, or within 2 km if COT < 1 (82599 matchup points for Sept-Oct 2013)
CBH [km]
Avg error (bias)
RMSE
Std of error r2
IDPS
0.7
2.7
2.6
0.452
Enterprise
0.3
1.8
0.760
Much better!

9. Sample matchup comparisons
The original IDPS with CLAVR-x input => Enterprise CBH
CloudSat Cloud Mask (gray) and VIIRS CTH/CBH (colored by cloud type)
Horizontal CBH contours
Enterprise CBH [km]

10. Ongoing work to provide an optimized CBH retrieval
For high thin Cirrus, CALIPSO-based extinction method by Yue Li (CIMSS) and Andy Heidinger (NOAA/STAR)
For deep convective clouds, CBH = Convective Condensation Level from NWP
The original version
for thin cirrus
The new version with the extinction method for high thin cirrus
Combined with the statistical method in the current CLAVR-x CBH routine

11. Nighttime CBH algorithm performance
New work in progress
Nighttime CBH algorithm performance
Preliminary results for the enterprise CBH and ARM SGP-C1 ceilometer data
(9 valid cases within a 1-km and 5-min matchup window in Jan-May 2015)
Within 2 km
NLCOMP or NWP products used for nighttime CWP
Ongoing work using ARM NAS (Barrow, Alaska) ceilometer data

12. Apply to geostationary satellites (I) GOES-W and GOES-E
CTH [km]
CBH [km]
Sample CTH and CBH from GOES-13 on 8 May 2015 (1815 UTC)
Similar CBH data (lower layer CBH) being produced from the NASA Langley Center

13. Apply to geostationary satellites (II) Himawari-8 AHI -> for the future GOES-R ABI
Sample CBHs at 0450 UTC on 25 March 2015
AHI 11µm TBs ( K)
CIRA CBH (km)
Cut the edge

14. For Cloud Cover Layer improvement
Once the optimized cloud base height estimate has been established and validated, we can use it for improved CCL products.
The cloud geometric thickness information allows for a pseudo-three-dimensional cloud field which can be used to estimate cloud fractions at lower levels below CTH.
Coupling the information with cloud classification and NWP temperature profiles would assist in providing useful parameters with regard to CCL retrievals.
Conceptual illustration of how cloud geometric thickness information can be used to modulate the layered cloud fraction (high/mid/low) by introducing additional cloud coverage at lower (unobserved via satellite) levels of the profile.

15. Satellite CBH of the upper-most layer and CCL
For potential users …
Satellite CBH of the upper-most layer and CCL
(as supplementary info)

16. Summary
Retrieving CBH is difficult. Our evaluation showed the IDPS CBH environmental data record provided only marginal skill.
Cloud Top Height and Cloud type errors significantly impact CBH.
CIRA developed a new statistical CBH algorithm constrained by CTH and CWP using CloudSat/CAIPSO and Aqua MODIS data. (Now part of the CLAVR-x system)
The enterprise CBH algorithm outperforms the other algorithms particularly when CTH is “within spec”.
Work in progress is exploring alternative fits for the optimized CBH retrievals such as a higher order polynomials to improve thick cloud base.
Validation efforts are ongoing for an extended CloudSat matchup period (Jan-May 2015) including nighttime CBH performance test and comparisons with CALIPSO for thin cirrus.
Once the optimized cloud base height estimate has been established and validated, we can leverage it to address forecaster needs for improved Cloud Cover Layer products.
Thank you!